HP 3457A teardown, some poblems + needs full calibration

[AMR Labs]

Got this 3457A relatively cheap off ebay, advertised as being fully functional, accurate voltage readings, and passing self test with a starting bid of $39. It sure sounded like the calibration should still be intact, so I tough this would be a good chance to get a nice 6.5 digit DMM as the lower cover was missing a chunk of plastic on one side towards the front where a rack mount bracket used to be attached and this might deter some of the other potential buyers. Fortunately in the end that damage actually looked much worse in picture then it actually was.

The unit arrived very well packed and is very clean inside, no signs of tampering, most chips are dated 1986, and just a few are 1985. I was hoping the internal battery would still be holding the calibration, but it was completely dead, like in zero volt. Looks to me that after the cover got damaged the unit was retired from service and put away into prolonged storage and never serviced again. I'm guessing this unit was actively being calibrated quite a bit, as a query of the CALNUM returned 121. This of course does not indicate 121 FULL calibrations, just all the individual parameter calibrations since the factory initialization. But at least gives an idea that the unit was probably being regularly calibrated perhaps even once a year until it was put away.

One thing I did not know, and of course found out too late, is that the Self Test OK message can still show even with some errors being detected, and the ERR annunciator on the display would start blinking (as explained on page 2-4 of the user manual), but of course this was unfortunately not captured on the seller images. This is very different from the 3468/3478 DMMs that will definitively not pass power up Self Test if the calibration constants have been lost. Definitively one thing to watch out if you plan on buying on of these off ebay. Anyway, upon checking the errors sure enough one of them (2 in total) was "9 Not Cal'd", and the second is just indicating "ACAL Req'd", which goes away by just performing an Autocal All.

The 3457A voltage readings when compared to my HP3478A  (has not be calibrated since at least 2014 when I got it with a good internal battery that I have replaced and had intact last calibration) where still within about 0.25% on all tested ranges, which of course does not really mean much as there is a one order of magnitude resolution difference between these two multimeters, but at least it shows its in the ballpark and seems to be working correctly. For 2W ohms I used my resistance test box build with 0.01% resistors (10R, 100R, 1K, 10K... 10M) and the difference with the '78A was much lower, only around 0.05% across all ranges except for the lowest 30Ohm range, obviously due to the resistance of the test leads interfering. But there is a problem in the highest 2W ohms range where it is showing about 267M ohm without any test leads connected. All other 2W ohms ranges appear to be working ok and readings are also in the general ballpark. I put a 30Mohm resistor (3x 10M 5% in series) in 2W mode and it reads about 26Mohm, which would roughly be the equivalent of having those 267M in parallel with it, so it definitively seems like something in the ohms section is causing this resistance leak. But in 4W mode this 30Mohm resistor can be measured correctly and it shows a reading of about 29 and change Mohms very much in line of what the 3478A reads.

As a first step build a small 3.0V stable reference with a MAX6126AASA30+T chip which provides 3.0V at 0.02% initial accuracy, and has 3PPM/C spec. Its the best I could find without a crazy price tag, and it is much better than the other alternative which was the somewhat cheaper REF5030 but with only 0.05% accuracy. Also the MAX chip I used only requires one external component (an output bypass capacitor), no trimming, and can be powered from anything between Vout+0.2 to 12.6V, so for added stability and low noise I used a rechargeable NIMM 9V battery. The 3457A reads this as 3.00009 from a cold start, and after about 60 minutes the reading stabilized at 3.00014. As a comparison, my 3478A was reading 3.00010 and settled to 3.00012 after about an hour or so. My 974A 4.5 digit portable DMM measured 3.0007 at both times, but I did not leave it on all the time so these would be just cold start readings. The calibration of the 3457A calls for a +/-0.0006% (+/-18uV) for the 3V range, but with only 0.02% accuracy of my present 3V standard I am currently at +/- 600uV. Not even close.

I have since replaced the internal 3V battery so any changes I made would be retained, also ran an Autocal All, and then using the MAX chip 3.0V reference calibrated that range and it seems to be holding the calibration and reading just fine after a couple of power up/downs. I will have to figure out how to calibrate the DC ranges and 2/4W Ohms which are the ones I would be mostly be interested in, but first need to see about the 267M open circuit reading in 2W mode. Any suggestions? I'm thinking it might be related to the Ohms input protection circuit, that involves 4 transistors (Q213-216), a Fet (Q212), and a diode (D201). From there on its basically into the Hybrid U101, which I just hope will not turn out to be the problem.

Anyway, here are a couple of pictures as well. Looking forward on getting some comments and suggestions.

[Kleinstein]

The ohms problem could be the ohms source (epsecially the transistors for the protection), but also the input bias of the normal voltage input.
So it would be a good idea to check the input bias.
This can be a high resistor (like 100 M or 10 M) for testing at zero, or a low loss / low leakage (e.g. PS or PP type) capacitor and looking at the drift rate.

[iJoseph2]

Is that a Schaffner IEC power input connector with Rifa caps inside?
If it is you should plan to replace it as soon as possible with a new one from a reputable supplier. Apparently NOS are no good.

[AMR Labs]

The ohms problem could be the ohms source (epsecially the transistors for the protection), but also the input bias of the normal voltage input.
So it would be a good idea to check the input bias.
This can be a high resistor (like 100 M or 10 M) for testing at zero, or a low loss / low leakage (e.g. PS or PP type) capacitor and looking at the drift rate.

Hi, do you mean the circuit around U111 and U112, Q111 and Q112 at the input amplifier? I am really not very familiar yet with the circuit, so not sure where that input bias is being injected. Also need to find another service manual, the one I have has the input section sliced into 3 separate pages in reverse order, and is a pain to follow.

Thanks.

[AMR Labs]

Is that a Schaffner IEC power input connector with Rifa caps inside?
If it is you should plan to replace it as soon as possible with a new one from a reputable supplier. Apparently NOS are no good.

Not sure yet if its the dreaded Shaffner, the writing is on the opposite side and will need to dismount to look at it. In any case its an item on the list. Thanks for pointing it out, I know they are a real stinker if they blow.

[AMR Labs]

Was reading through the detailed theory of operation and it occurred to me to place another voltmeter (my 974A) across the Hi/Lo inputs of the 3457A when in 2W resistance. The 974A reads 1.14x-1.15xV jumping around a bit on the last (x) digit, and the 3457A seems to correctly measure the input impedance of the 974A and shows around 10.7Mohm. When I switch to 4W resistance the 974A then shows 6.866V and very stable.

Not sure this helps in any way but out of curiosity wanted to measure it, and perhaps someone can verify this on their own 3457A.

[AMR Labs]

Found this site that has some detailed schematic with some additional annotations:
https://xdevs.com/article/hp3457a/

Particularly found this image very useful:
https://xdevs.com/doc/HP_Agilent_Keysight/3457A/img/8-77.png

It seems to indicate that the Ohms Current Source sends 8V towards the input High terminal (from IC213 pins 6 and 2) by going first into pin 18 of Hybrid U101, then leaving through pin 15, on to Q213-Q216, then through L201/CR201 and a couple of relay contacts. Would this be correct?

If 8V where supposed to be present at the input, then a reading of about 267Mohm would imply a current of just a bit under 30nA. I know the Ohms current source determines the amount of constant current sent to the input terminals, not the voltage per se. But I wonder what voltage is supposed to reach there, the 8V indicated in the schematic? And why am I only getting 1.14V there.

[Kleinstein]

The current path is about right, but the P-MOSFET inside the hybid between the pins 17 and 15 is not operated as a swtich, but as a linear elemet. It is the one to control the current to a constant level.  So the voltage at pin 17 is regulated to be at 8 V, but the voltage at pin 15 is lower, depending on the DUT.

The current is set by the difference of the 8 V level at the U213 inputs and the 12 / 8.4 V voltlage at the output of U211 / inputs of U212.
The resistors are the 4 K / 40K / 400K / 4 M connected to Pin17.

The lowest current ( 30 M range) should be 0.4 V divides by 4 M and thus 100 nA.

As there is some drop in the diode and protection the maximum voltage at the ouput shuld be at some 5.5 V up to a bit over 6 V at very low current. This should be enough to get the maximim of the 3 V range as the highest votlage to measure with high impedance.

The usual test is to check if the tests currents are about right (can still be a few percent of nominal).

The very high resistors > 30 M would be measured with the help of the 10 M input divider (in parallel to the DUT) and not the classic mode with the constant current source. This is relatively sensitive to the divider and thus needs a careful calibration. It may be just an extra cal step missing.

[AMR Labs]

I will start by saying that the "phantom" 267M reading with the terminals open has been solved. Let me just enumerate all that was done that both did nothing to contribute to the solution, and what did.

Last night I took a whole bunch of readings around all the input elements writing down all the findings, and read the detailed theory of operation (several times) to try and get a bit more understanding on how the inputs are being processed, and perhaps finding a voltage that is off, or something missing. I even removed the rear inputs completely and even the AC board and put it aside. Needless to say the problem was still there. BTW there was a false positive at some point, in that while measuring the voltage at pin 18 of U101 (same node as U213 pin 3) I saw in the corner of my eye GOHM on the display on the 300M range. But this was short lived, as it turns out with open inputs the indication was rather 0.0000 GOHM, and not the correct OVLD GOHM. But at least got to see the "GOHM" text on the display, so that was at least a bit uplifting. I also noticed that this would even happen with the black return lead of the external DMM disconnected and only the red lead at pin 18. So I assume some noise was being introduced that was somehow masking whatever was causing the 267M phantom reading.

Anyway, I could not find anything wrong with any of the measurements other than some of them did not seem right specially for the 30/300M ranges but they where clearly being affected by the portable external DMM I was using (the 974A) that "only" has a 10M input Z. So next tried using the 3.0V range of the 3478A (BTW I also have a 3468A) which has 10^10 ohms input Z and it worked better in a couple of spots but in others it would just cycle about every second between reading the voltage (about 1.1V) and then showing overload as I had it in manual range so it would actually stay in the 3V range I then tried also with the 3468A but got exactly the same result. I was puzzled by this because the portable DMM showed there was just 1.1V at the node with a perfectly stable reading (only loaded down by its lower internal Z), no pulses no variations. I even put a scope on the node and of course there was nothing but DC there. So I assume some kind of common mode current was creeping between the measurement point in the 3457A and the 3468/3478. I did check the ripple on all 3 DC busses +15, -15, and +5 and all looked very clean.

While trying to understand further some of the elements I started measuring voltages around U214 the Harris analog switches. Something did not seem right as in the schematic the terminal of pins 3 and 4 had very different voltages (15v and 4V) and yet that switch is shown as closed in the schematic, and voltages where equal between switch pins 1 and 16 (both 4V), even tough that switch is shown as open in the schematic. And furthermore the 5V steering logic of the switches seemed to be stuck on Hi all the time. I did not have any external resistor connected to the input terminals so later found out that when in OVLD condition the steering logic only start switching states when there is actually some resistance connected to the input to be measured. Also found the datasheet of the Harris  HI3-0390-5 chip, and was able to understand how it works and everything fell in place and it was just that the schematic is actually showing the switches in the opposite position to being in the 2W OVLD condition which is what I had at the time. Long story short, there was nothing wrong the IC214. But at least I was able to fully understand its operation and function. Also among others also understood how the ohms current source works by using the 4 current setting resistors inside U101, and about the voltage difference of 4V (12V-8V) to generate the test current for all but the 30/300M ranges, for which only 0.4V (8.4V-8V) are used instead.

I basically went on like this with the other areas of the input signal conditioning section and determined by reading the output voltage going to the ADC while connecting some external resistors to the input and measuring the resulting voltages it showed that all ranges except the 300M where behaving correctly, and only the 300M input to the ADC seemed wrong, so at the very least I knew the problem was definitively upstream from the ADC. Anyway, it was then close to 3am so went to bed and let the ideas churn around a bit and start fresh in the morning.

So next morning one of the first things that popped into my mind is what if the 300M range problem showing 267M with no input is just related to the unit being uncalibrated. Why not attempt a Volt and 2/4W calibration with values as close as possible just to get through the procedure notwithstanding any inaccuracies which at this point would not really be that important.

Started by doing all the preliminary calibration procedures: run a self test, which resulted in a hardware error, which of course was due to the AC board being on the other end of the table. I then connected it back and the self-test went OK this time. Now on to running an Autocal routine which also went without any issues.

So let me just quickly check again the 300M range, nope, still have the same problem, its not  going to be that  easy. So on with the next calibration procedures: DC Offset cal. I do have two heavy #6 solid copper wire U shaped jumpers that are ideal. All ranges starting with 300V and down calibrated fine to zero.

Now volts gain calibration. For this step I used a hi-volt capacitor restorer that outputs 150-600V, and trimmed it as close as possible for 300V with a variac and calibrated the 300V range. Then used my variable power supply for the 30V range, the MAX chip voltage standard I build for the 3V range, and then used a 5K 10T potentiometer fed from the 3V standard to dial in 300mV and 30mV to cal those two remaining ranges. All went well, no issues at all. Then did the 3V linearity calibration by reversing the polarity as indicated in the manual, and that too went well.

Maybe 300M range working now? No.

Next calibrate 2W ohms. For this I used pairs of soldered together 1/4w 5% resistors (3x 10M for the 30M "standard") just to get into the ballpark of full scale for all the ranges, and checked them with the 974A and wrote down the average values. Most where slightly under like 2.969M, 295.150K, 29.839K, 2.9675K, 297.55R, and 29.66R. Only the 30M string was slightly above FS, at 30.050M. I started calibrating all ranges starting with the 30M and down, and all went well.

Time to check again on that 300M range indication, and YES, it now showing "OVLD GOHM" with nothing attached to the inputs. Success!

Anyway I though while I'm here, let me also do the 4W ohms offset and gain calibrations as well. I do have two of those heavy copper U shaped jumpers, and both where connected together with another short piece of wire and all ranges starting with 300M and down where successfully set to zero. Then did the gain calibration of all ranges, again starting from the top 300M and down, and again all good.

Looked again at the 300M 2W indication, and yes, its still good indicating OVLD GOHM.

So guess at this point I have a "fake" calibration of the volts and 2/4W ranges, and that definitively solved the problem, which really was never a problem, just lack of calibration. Hope this helps someone in the future encountering a similar situation. Time to button up the 3457A, and start thinking how to get it properly calibrated. Unfortunately where I live there are no calibration facilities, none. I am outside of the US, and my "closest" facility would probably be in Miami FL, or perhaps Puerto Rico, but I am guessing that Fedexing this instrument would cost about the same to either destination and I feel more confident about mainland USA for a decent calibration which I still don't even know how much it is going to cost. My other alternative would be to get a calibrator to me, but I am not prepared to spend hundreds (or thousands) of $$$ on something that matches the calibration accuracy requirements of this DMM. I really only would need hobby level accuracy, so i was thinking that another possibility  would be to build a DC calibrator that would at least include 30 and 300mV, 3V and perhaps also 30V, which I anticipate would be the ranges I would use most. And then get from Mouser or Digikey a set of resistors with best possible accuracy and values to calibrate the 2/4W ohms ranges. And then send both the voltage "standards" and resistors off to someone with a 3458 or Fluke 732 to just characterize the values for me with documented environmental conditions so I could hopefully replicate them as closely as possible, and then send them back to me. I would of course be happy to pay for all shipping and cover time involved up to a reasonable amount. In general I am happy with my 3468A and 3478A performing quite adequately, but once you get to the level of a 3457A, well it would really be an injustice to just get it ball parked, it really deserves the best possible and viable calibration given the logistical limitations. I wish the backup battery had not died who knows when and calibration was lost, as intact calibration constants from the last good calibration procedure would have probably been enough for me, at least for a while.

Sorry for the long post, but I had a quite bit I wanted to document, so thanks for reading if you made it this far.

EDIT: added 2x pictures.
EDIT2: couple of spelling corrections

[AMR Labs]

The current path is about right, but the P-MOSFET inside the hybid between the pins 17 and 15 is not operated as a swtich, but as a linear elemet. It is the one to control the current to a constant level.  So the voltage at pin 17 is regulated to be at 8 V, but the voltage at pin 15 is lower, depending on the DUT.

The current is set by the difference of the 8 V level at the U213 inputs and the 12 / 8.4 V voltlage at the output of U211 / inputs of U212.
The resistors are the 4 K / 40K / 400K / 4 M connected to Pin17.

The lowest current ( 30 M range) should be 0.4 V divides by 4 M and thus 100 nA.

As there is some drop in the diode and protection the maximum voltage at the ouput shuld be at some 5.5 V up to a bit over 6 V at very low current. This should be enough to get the maximim of the 3 V range as the highest votlage to measure with high impedance.

The usual test is to check if the tests currents are about right (can still be a few percent of nominal).

The very high resistors > 30 M would be measured with the help of the 10 M input divider (in parallel to the DUT) and not the classic mode with the constant current source. This is relatively sensitive to the divider and thus needs a careful calibration. It may be just an extra cal step missing.

Yes, this is one of the things I started understand once I was reading through the detailed theory of operation of the input signal conditioning and ohms current source sections, and by taking voltage readings (or trying to) at the corresponding nodes. Thanks very much for pointing out this information, it confirms I was understanding things correctly. Your input is greatly appreciated.

[Kleinstein]

The DC and resistance calibration is the easy part. The current gets more tricky and the AC part even more.
I have not read through the cal instructions there, but at least a step with a square wave to get the divider compensation right may be woth the effort.

The 3457, 3478 and 3468 are all 3 with the slightly unusual 3 V range as there prime range. Other meters tend to be more like 2 V or 10 V and the relatively cheap cal sources form China are usually made for these ranges.

Shipping the 3457 with the already damaged case is probably not ideal.

A reasonable stable DIY 3 V and 300 mV (maybe 30 V too) source to send out sounds like a viable idea, especially with 3 meters that use similar ranges.
For the 300 V and 30 V to 3 V step, one could consider a DIY Hammon divider. This gives a resonable accurate 1:10:100 divider.

[AMR Labs]

I don't really anticipate much use for the AC functions, and yes those would be much more complicated to calibrate with a self build calibration source. And by the way no 60Hz  because the manual specifies that the all calibration voltages need to be 1KHz. Imagine a source of 300V at that frequency, that should be fun and "easy" to build. I would think that I could get the 3 lower ranges up to 3V directly calibrated off my audio oscillator, and the 30V range by running the signal through an audio amplifier, or even just a plain transformer might do it.

But on the other hand the accuracy of the AC ranges of the 3457A are anything but stellar, rather lagging behind its DC performance. The specified calibration sources for these ranges are only required to have 0.1% accuracy, so that alone says a lot. Nor would I anticipate having much need for the very limited AC or DC current measurement capabilities. However I will definitively cal the frequency range, I can use my 10MHz Rubidium oscillator and divide it by 10 with a simple 7490 chip, or use a direct output from my Motorola radio service monitor which has the high stability ovenized master oscillator option build in. Not sure how useful that frequency range on a Desktop DMM really will be, as I have other stand alone counters but I'd guess it would be nice to have it properly calibrated just in case. Not sure if it uses reciprocal counting and be able to read low audio frequencies without involving a very long gate time, if so that would be a plus.

I am planing on building at least a 30mV and 300mV sources that would be using the 3V source, just a simple 2 resistor attenuator with an added 25 turn quality trimmer to fine tune, and then going into a non-inverting op-amp buffer for isolation and low impedance output should do it. For this I am considering to use the ICL7650SCP (spec sheet bellow) which seems suitable for this purpose given its very low drift and offset and no trimming required. Its an old part but one can still find them on ebay from reliable US sources. For the 30V source perhaps build a voltage regulator stage that also uses the 3V source as a reference. Not so sure about how to build a reliable source for the 300V range, otherwise I'll just comparison calibrate best as possible against the 3478A with a HV power supply I build to restore electrolytic capacitors, and that provides 150-600V, and I could as I did once before use a variac to trim the voltage and do it late at night to try and minimize line voltage fluctuations and disturbances.

About the hole in the lower shell of the case, it really does not compromise the structural integrity of the instrument and it still feels very solid as the affected area is really rather a very small portion of the affected part. Nor does it expose much of the innards, other than the back side of input connectors. I will eventually buy a HP3488 control unit that uses those exact same covers and they can usually be found in good shape for as low as $25 on ebay all the time, as it seems these where sold in the thousands during their time and presently there is not much use for them anymore. Thing is the shipping due to its size will cost some to get it to me, although it would not bee too bad. But for now putting it off, waiting to have a bit more justification to invest in aesthetics. For now just to prevent humidity or dust to get in, I simply sealed off the hole with some blue 3M tape that seals and sticks very well, but will not leave any gummy residue when removed.

Will post any updates on the voltage sources soon as they get build.

EDIT: correct spell.

[25 CPS]

I don't really anticipate much use for the AC functions, and yes those would be much more complicated to calibrate with a self build calibration source. And by the way no 60Hz  because the manual specifies that the all calibration voltages need to be 1KHz. Imagine a source of 300V at that frequency, that should be fun and "easy" to build. I would think that I could get the 3 lower ranges up to 3V directly calibrated off my audio oscillator, and the 30V range by running the signal through an audio amplifier, or even just a plain transformer might do it.

1 kHz at 300 volts...  If I had to do it quick and dirty, I'd get an output transformer from a tube amp and backfeed it signal into the secondary windings that are intended to drive a speaker and adjust the amplitude of the input signal until measuring 300 volts off the high voltage primary winding on a known good meter then patch in to the 3457 getting the quick and dirty calibration.

[Kleinstein]

The ICL7650 is old and OK to work with. There are modern alternatives (e.g. max4238, AD8628, MCP6V76) that don't need the extra caps - though they usually only come in smaller SMD case.
The adjustment usually does not need exact 300 mV or excat 30 mV, so no real need for an extra trimmer. Just measure the actual ratio or get the divided voltage also measured. The trimmer is OK if used only for a small adjustment in the right way. Trimmers tend to be very drifty.

For the 30 V part I would also consider using the 3 V ref. and than an amplifier for the step up to 30 V this is still in the range of some OPs and thus relatively easy. One may be able to do a consistance check with the 300 mV as the starting point.

[AMR Labs]

1 kHz at 300 volts...  If I had to do it quick and dirty, I'd get an output transformer from a tube amp and backfeed it signal into the secondary windings that are intended to drive a speaker and adjust the amplitude of the input signal until measuring 300 volts off the high voltage primary winding on a known good meter then patch in to the 3457 getting the quick and dirty calibration.

There you go, certainly another alternative I was kind of also thinking about when considering to get the 30V AC source from a generator, but it could certainly also be used for 300V with the appropriate turns ratio transformer and enough input into the output secondary, probably through an amplifier as its output would perfectly match the low Z side of the transformer. One of those 5K or 10K to 4/8 ohm output transformers should do it. Thanks for your contribution.

[Kleinstein]

At only 1 kHz many of the normal mains transformers (e.g. 3 VA range)  are still good enough. There are a little more eddy current losses, but nothing dramatic. A transformer can also provide stable ratios. So at least the half way point from a split winding should be pretty accurate as they usually really get the counts right to allow parallel connection.

[AMR Labs]

The ICL7650 is old and OK to work with. There are modern alternatives (e.g. max4238, AD8628, MCP6V76) that don't need the extra caps - though they usually only come in smaller SMD case. The adjustment usually does not need exact 300 mV or excat 30 mV, so no real need for an extra trimmer. Just measure the actual ratio or get the divided voltage also measured. The trimmer is OK if used only for a small adjustment in the right way. Trimmers tend to be very drifty.

That is correct, as long as the voltage source output falls anywhere between 2.95 to 3.03 it should be fine for the 3457A to accept the calibration data. I decided to use the ICL7650 because I saw it being used for a similar application, and was able to source them relatively easily. I just build the 300mV source on a protoboard and it seems stable short term, and output with zero input was 1.65mV, open or shorted to ground.  So far there has been some gradual downwards drift in the 10s of uV but I'm waiting to see if it will settle or slow down at a certain point, so will be watching it for a couple of hours. It will most probably get more stable once I properly build the circuit on a PCB. Still I am also going to look into those other newer Op-amp alternatives that you are pointing out. Not worried about them not being DIP, as a matter of fact the MAX chip for the 3V source I had to build on a SIP to DIP adapter, so no problems there.

For the 30 V part I would also consider using the 3 V ref. and than an amplifier for the step up to 30 V this is still in the range of some OPs and thus relatively easy. One may be able to do a consistance check with the 300 mV as the starting point.

What would you recommend to use for the 30V amplifier? I'm guessing it would have to be a precision Op-amp that can take at least a 35V single ended power rail. One with 45V would probably be even better so as not to push the required output too close to the upper limit and get somewhat better headroom and possibly stability as well by keeping the Op-amp in a more linear region. I'm also going to look myself with the Mouser.com parametric search see what they can offer.

[Kleinstein]

For the 30 V amplifier I would consider a low voltage AZ OP and than a 2nd OP like OPA991 for most of the gain inside the control loop (as a compound amplifier).  The supply could be something like -1 V and + 33 V. The OPA991 is rail to rail out and OK for up to 40 V, so there is some headroom. The final loop closure and thus precision comes from the AZ OP. So no need to avoid the rails very much.  Other OPs for the "high voltage" stage are OK too as the voltage of interrest would be 3 V and 30 V.
The input to the x10 gain stage would be either the 3 V ref or the 300 mV level to get a self constency check. 300 mV in and x 10 gain should give back 3V.

One could also do it with a single higher voltage AZ OP, like LTC2057 or MCP6V51.

It can help to have some 100 Ohms at the outputs to avoid capacitive load to the OPs - some DMMs have enough capacitance at the input to bring some OP to the edge of oscillation.

An auxiliary -0.5 to 1 V negative supply can be generated with a simple diode drop on the negative supply input, just make shure there is more current from the positive side.

[AMR Labs]

OK great information, thank you.

Will be ordering from Mouser, a couple of 98.8K, 9.98K, 1.11K and 1.01K 0.1% for the 30mV and 300mV voltage dividers that should hopefully fall very close on the exact expected output and so exclude the need of a trimmer. But just in case those combinations go a bit over and since I don't have much calibration up margin on the 3457A as it wants to switch range at 3.03 FS, also ordering 100K and 10K as the top resistor instead, so the output should fall a bit lower around 299.72 and 29.99mV.

Also since they are all just about a dollar and change each, will order one each of your suggested Op-amps: OPA991, LTC2057, and MCP6V51T so I could compare them.
And need to get a SOT-23-6 to DIP-8 converter board (for the OPA991), probably off ebay.

Anything else I might be forgetting?

[Kleinstein]

One can get a 1:10 divider with 3 equal resistors in series and 3 equal resistors in parallel. The idea is used in the Hammon divider to a quite accurate 1:10 division with a few adjustment / test steps. 2 switches are used to switch the 3 resistors in paralel between a parallel and series configuration. The series coonfiguration is used to check that the 2 sets of 3 are close to each other and if needed trim 1 resistor slightly. With the parallel configuration one than gets a quite accurate divider, even if the 3 resistors inside the sets do not match perfectly.

http://conradhoffman.com/mini_metro_lab.html
This link may give some ideas on how to build and use jumpers as switches.

The 3 in series and 3 in parallel trick may also work for checking the resistor ranges relative to each other: 3 equal resistors of 30 K give 90 K in series and 10 K when in parallel. The ratio of the 2 values is quite accurate at 1:9. The error goes with the relative tolerances squared + switch resistance. So starting with 1 % reistors the ratio would still be accurate to the 0.01% range (maybe an extra factor of 2 or 3 - I don't remember). Start with slightly better resistors and you get pretty good.

If really needed a set of 10 equal resistors could in a similar way be used to cover values in the 1:10:100 range as: 10 in parallel, 3 S*3P, and 10 in series.


Resistor networks with equal resistors have quite good matching to start with. Something like t
he NOMCA / NOMCT series may be an alternative to 0.1% accuracy resistors.  SO16 is still managable as dead bug.

The MCP6V51 is especially lower cost than the LTC2057, but with more current noise . Current noise specs on AZ OPs are not allways reliable. Microchip has a slightly different definition of bias (average of pos and neg side input current, even if they compensate). So you have to add something like 1/2 the offset current to get comparable numbers with others.

Another interesting test to do with the meters would be to check the DMM input bias at different voltages. This can be done with a low loss (PP or PS type) cap of some 10 nF. So it may be worth getting such cap if you don't have.

[AMR Labs]

http://conradhoffman.com/mini_metro_lab.html
This link may give some ideas on how to build and use jumpers as switches.

Interesting link, and website as a whole. Will definitively look into the article from Electronics Now. There is also another article on the website called "Build a 0.1X and 0.01X Hamon Divider", that sounds perfect for the 300mV and 30mV sources. The article reminded me that as a kid I used to read Popular Electronics (that at some point absorbed Electronics Now) every month and built quite a few of the published projects since the 70s all the way until its sad demise in the early 2000s. And of course the once or twice yearly EEH editions (Electronic Experimenter's Handbook) where always greatly enjoyed as well, as they where a compilation of the best Popular Electronics projects from a year or two before.

BTW, all issues of Popular Electronics and many more other electronics (and other types) of older magazines have been preserved in PDF and are available for free download from this excellent historical website:

https://worldradiohistory.com/Popular-Electronics-Guide.htm

Another interesting test to do with the meters would be to check the DMM input bias at different voltages. This can be done with a low loss (PP or PS type) cap of some 10 nF. So it may be worth getting such cap if you don't have.

Not sure how one would do this with a cap, can you please explain what the procedure is? I am sure I do have PP (PolyPropilene?) caps of that value. But please refresh my mind what does the "PS" stand for?

Thanks.

[Kleinstein]

PP stands for polypropylen ,PS for polystyrene (older type of caps, but very low leakage).

The idea to measure the leakage / bias current is to watch the curve / speed how the capacitor charges from the current. The meter itself can be used to measure the voltage. A 10 nF cap is more common than resistors in the GOhms range. Addition one can measure the input current at different voltages, not just at zero input.

With 10 nF a current of 1 pA would give a rate of change of 0.1 mV per second, which is easy to measure.  It is not so much about high accuracy, mainly about the rough range. The accuracy is anyway limited by extra capacitance inside the meter and most caps are more like 10 % tolerance, but that is alread OK in pA range.

[AMR Labs]

Well finally got some time to play a bit more with the 300mV and eventually also the 30mV and 30V sources. Also had received a few 0.1% axial resistors and the new zero drift op-amps from Mouser some time ago but too many other projects in the way, and a bit of laziness.

I have to say that the 300mV I breadboarded around the ICL7650CPS Opamps I got from ebay where never terribly stable, even with the input voltage divider changed to Vishay 0.1% metal film resistors. Initial reading at the output through a 100 ohm resistor measured 299.69mV on the 3478A. But 2 hours later was at 298.45mV, so a drift of about 1.28mV. Not sure if that is the best this older generation supposedly stable Opamp can do on a breadboard. Also the third decimal point of the reading was really too unstable to include. Then again I had not warmed up the 3478A but  that 1uV digit position was still too jumpy after the 2 hours between readings had passed.

Anyway I now used the OPA991C Opamp which I chose in the SOT23-6 case and mounted it on a DIP adapter. With the input attenuator from 3V with a 10K and 1.11K resistors the output reading so far as been rock steady down to 3 decimals which is the most I can get out of the 3478A in the 300mV scale. I read 299.545mV when I first turned it on, and so far over an hour later it has only drifted up 1uV (to 299.546mV) although it is occasionally moving between 545-546-547. Also the third digit uV decimal is very stable so actually it now makes sense to include it. The Opamp is being powered with a single ended +15V supply and ground. Will leave this powered on for a few more hours and see how the output stability behaves.

Later will also try for the 30mV source by changing the input attenuator resistors to 100K and 1.01K respectively, and then the 30V by increasing the Opamp supply voltage to 40V and adding 10x gain. Just realize that I forgot to order a stable feedback resistors for this application but at least will give it a try with what I have on hand now to see how it behaves.

[Kleinstein]

The ICL7650 is normally very stable. However it needs the external caps and there it may be effected by the breadboard. I have never used the ICL7650 on a breadboard - I have used other AZ OPs (e.g. MCP6V51) with an adapter baord and they were quite stable / low noise. With some time to equilibrate the breadboard may not be that bad.

[AMR Labs]

Since about 4 hours ago, including a change of ambient temperature from 27 to 30c, and humidity from 50% to 75% (I turned off the A/C), the output of the 300mV source has gone up only by 5uV, from 299.546 to 299.551mV. Guess that is not too bad specially for a breadboarded open air circuit. I am starting to like the OPA991 and will order a few more for the other voltage sources.

Not sure what happened to the ICL-7650, and yes I did include the required caps. Maybe it was the 0.1uF caps I used (axial multilayer) that caused the instability but I doubt it. Or perhaps it was too good to be true getting 10 of these from China for total $11.50 Standard SpeedPAK shipping included. The chips do not look fake, relabeled or repainted, but who knows these days, they are only getting better in passing counterfeit parts. At the time there where no other reasonable priced sources.

For the 30V source in non-inverting mode I figure to use 1K for the input resistance and 8.97K for the feedback resistor. That combination should provide a gain of 9.98 times and with 3V input should provide and output of just a bit under the DMM FS range spec at 29.940V.

You said it would be desirable to feed a small negative voltage to the minus rail of the Opamp with an "auxiliary -0.5 to 1 V negative supply" that can be generated with a simple diode drop on the negative supply. So basically connect the negative rail to ground through a diode instead of directly?